The Agent Skills Directory

[METADATA_POISONING]: The skill contains a file named skill-report.json that mimics a legitimate security audit. It falsely claims a safe status and references non-existent AI models such as 'gpt-5.5', which is a deceptive practice intended to bypass security review workflows.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external PDF files using ingestion points like PdfReader and pdfplumber. This exposes the agent to indirect prompt injection, where malicious instructions hidden in a document could attempt to subvert the agent's logic during extraction or form completion.\n- [COMMAND_EXECUTION]: Documentation and scripts suggest the use of external shell commands for PDF utilities including qpdf, pdftk, and pdftotext to perform document merging and layout-preserved text extraction.\n- [DYNAMIC_EXECUTION]: The scripts/fill_fillable_fields.py script performs runtime monkeypatching of the pypdf library's get_inherited method. While used to fix a specific formatting bug, this involves dynamic modification of executable code paths at runtime.

pdf-analyze